竞价一小时童年 · AI 与养育产业 · 2026-07An auction on an hour of childhood · AI & the parenting industry · Jul 2026

家长正用最先进的 AI 全副武装孩子，去赢一场 AI 正在取消的比赛——这个行业是一座竞价「一小时童年」的拍卖行 Parents are arming their kids with cutting-edge AI to win a race that AI is busy cancelling — this industry is an auction house bidding on 'one hour of childhood'

军备竞赛押注的技能——刷题、标准化应试、少儿编程、文书润色——恰恰是大模型最先、最擅长替代的；而 AI 难替代的体能、审美、社交、心理韧性与「想学」的内生动机，恰恰是被应试挤压掉的。The skills the arms race bets on — drilling, standardized tests, kids' coding, essay polishing — are exactly what LLMs replace first and best; while what AI can't replace — fitness, taste, social skill, resilience, and the intrinsic will to learn — is exactly what the testing squeezes out.

家长退化为焦虑的「项目经理」：把孩子当「项目」运营，核心职责＝分配预算（家庭资金）＋排期（孩子不可再生的时间）。AI 没有改变拍卖的标的，只是把供给边际成本压到接近零——让拍卖更密集、更早开始、更难退出。The parent degrades into an anxious 'project manager': running the child as a 'project,' allocating budget (family money) and schedule (the child's non-renewable time). AI didn't change what's being auctioned — it drove the supply-side marginal cost toward zero, making the auction denser, earlier, and harder to exit.

最前面压一块诚实层「底层数学」（遗传方差天花板、位置竞争囚徒困境、杠杆排序）；主脊是养育生命周期八节点（备孕→就业），每节点标注传统 vs AI + 冲击强弱；判断层用「便宜 vs 稀缺」双栏把不对称讲透，右侧立六块硬骨头。这是一张批判性行业解剖，不是育儿或投资建议。相邻议题见姊妹图：应试 / 考试筛选→edu、把人当系统优化→mind、注意力捕获→addiction。 Up front sits an honesty layer, 'the honest math' (the genetic variance ceiling, positional-competition prisoner's dilemma, leverage order); the spine is the eight-node parenting lifecycle (pre-conception → employment), each tagged traditional vs AI + impact; the judgment layer uses a 'cheap vs scarce' two-column to lay bare the asymmetry, with six hard bones on the side. A critical industry dissection, not parenting or investment advice. Adjacent topics on siblings: exams / test-selection → edu, optimizing the self as a system → mind, attention capture → addiction.

传统节点Traditional

已商品化 · AI 压价Commoditized · cheap

被抬价 · 塑造人Re-priced up · scarce

军备竞赛 · 焦虑Arms race · anxiety

底层数学 · 遗传天花板The honest math

58–62%

学业成就遗传度（双生子研究 TEDS/Shakeshaft）；智力遗传率随龄从婴儿约 20% 升到成年约 60%，而共享环境（家庭+学校）对认知的影响在成年后趋近于零The heritability of academic achievement (twin studies TEDS/Shakeshaft); intelligence heritability rises with age from ~20% in infancy to ~60% in adulthood, while shared environment (home+school) approaches zero for adult cognition

97%

好未来九章大模型 K12 数学「文字题」正确率（⚠️厂商自述）——军备竞赛押注的标准化应试，正是大模型最先替代的技能TAL's Jiuzhang model's accuracy on K12 math 'word problems' (⚠️vendor claim) — the standardized testing the arms race bets on is exactly what LLMs replace first

¥190.6亿

中国 AI 学习机 2024 销售额（同比 +37.6%、销量 592.3 万台），拍卖场从线下门店搬进硬件与 App，边际成本压到接近零China's AI study-tablet 2024 revenue (+37.6% YoY, 5.92M units) — the auction moved from storefronts into hardware and apps, marginal cost near zero

−25亿

少儿编程龙头童程童美累计亏损、2025 大面积闭店（221 家直营/18.26 万学员）——押注「未来必备技能」，恰是 AI 最先取消的赛道Kids-coding leader VTron's cumulative loss, mass closures in 2025 (221 stores / 182.6k students) — betting on a 'must-have future skill' that AI cancelled first

口径警告：本页是批判性行业分析，非育儿 / 教育消费 / 投资建议。遗传度数字须按口径读：学业成就约 58–62%（双生子法）、SNP/GWAS 约 30%（12–16%）、智力随龄 20%→60%——不同能力 / 年龄 / 方法，勿混用；「共享环境 20–31%」是童年学业静态解释力，「成年趋近零」是成年认知纵向趋势，口径不同不矛盾。补习效应量并陈：小样本中等（d≈0.42–0.67），但规模化后缩到 1/3–1/2。凡「九章 97%（限 K12 数学文字题）、成本降 800–900 倍、各家销量第一、小思 2.3 亿次唤醒、火火兔交互 5 倍、ApplyBoard 95% 成功率」等厂商自述 / 预测打 D 级；松鼠 AI 提分类数字无独立验证一律 D。每张卡片右上角 A/B/C/D＝证据强度。 Basis warning: a critical industry analysis, not parenting / education-spending / investment advice. Read heritability by its basis: academic achievement ~58–62% (twin method), SNP/GWAS ~30% (12–16%), intelligence rising 20%→60% with age — different abilities / ages / methods, don't mix; 'shared environment 20–31%' is childhood-academic static variance, 'approaches zero' is the adult-cognition longitudinal trend — different bases, not contradictory. Tutoring effect sizes shown together: moderate in small samples (d≈0.42–0.67) but shrinking to 1/3–1/2 at scale. 'Jiuzhang 97% (K12 math word problems only), cost down 800–900×, various sales-leader claims, 230M wake-ups, 5× engagement, ApplyBoard 95% success' are vendor claims / forecasts, grade D; Squirrel AI score-lift figures lack independent verification, all D. Each card's top-right A/B/C/D = evidence strength.

◆ 诚实层 · 底层数学（最重）The honesty layer · the honest math

在竞价之前，先接受三个残酷的数字Before you bid, accept three brutal numbers

这是整张图区别于「产品目录」的关键。在 AI 大幅降低知识获取成本的今天，家庭的教育焦虑与投入却不降反升——因为这套底层数学从未改变：基因方差有天花板、补习规模化后缩水、位置竞争是零和的囚徒困境。This is what separates the map from a 'product catalog.' Even as AI slashes the cost of getting knowledge, family anxiety and spending keep rising — because the underlying math never changed: a genetic variance ceiling, tutoring that shrinks at scale, and positional competition that is a zero-sum prisoner's dilemma.

学业成就的方差分解（双生子法）Variance of academic achievement (twin method)Shakeshaft 2013 · A

遗传 58–62%Genes 58–62% 共享环境 ~36%Shared ~36% 非共享Non-shared

智力遗传率随龄上升（共享环境成年趋近零）Intelligence heritability rises with age (shared → ~0)Plomin & Deary 2015 · A

婴儿 ~20%Infant ~20% 青春期 ~40%Teen ~40% 成年 ~60%Adult ~60%

Scarr-Rowe 假说：对高 SES（优势）家庭，充足资源已让基因潜能充分表达，学业差异主要由基因方差主导——靠堆砌「过量」教育资源 + AI 产品去突破孩子的基因天花板，转化率极低、资源浪费。共享环境对成就的补偿作用，主要发生在资源匮乏的低 SES 家庭。The Scarr-Rowe hypothesis: in high-SES (advantaged) families, ample resources already let genetic potential express fully, so differences are driven by genetic variance — piling on 'excess' resources and AI products to break a child's genetic ceiling has a very low conversion rate and wastes resources. Shared environment's compensating effect mainly appears in resource-poor, low-SES families.

▸ 位置竞争 · 囚徒困境Positional competition · the dilemma

所有人加码，只让钱沉淀给机构Everyone bids more; only the vendors win

高考 / 名校录取是零和博弈，回报由相对排名而非绝对分数决定。「别人都在补」的剧场效应下，单方退出＝丧失阶层流动 → 教育投入恶性通胀。AI 降低优质辅导门槛，只会推高整体及格线、让所有家庭同时加码、相对位置不变。金句：「很多补课不是买增长，而是买不掉队的幻觉。」College / elite admission is a zero-sum game; the payoff is set by relative rank, not absolute score. Under the 'everyone's tutoring' theater effect, unilateral exit means losing mobility → spending inflates malignantly. AI lowering the bar to quality tutoring just raises the passing line, so all families bid up and relative positions stay put. As one report puts it: 'many tutoring purchases don't buy progress; they buy the illusion of not falling behind.'

▸ 影子教育 · 边际递减Shadow education · diminishing returns

补习效应，规模化后缩到 1/3–1/2Tutoring's effect shrinks to 1/3–1/2 at scale

对资源薄弱儿童，增加辅导显著提分；但对已有较多资源的阶层，边际回报极小、甚至与公共教育「挤出」。证据并陈：小样本中等效应（d≈0.42–0.67），但 2024 年 265 项 RCT 元分析显示项目扩大到政策相关规模时，效应量仅为小样本的 1/3–1/2；NBER 约 0.37 SD。UNESCO 已把私人补习定义为全球性「影子教育」。For resource-poor children, more tutoring lifts scores markedly; but for resource-rich strata the marginal return is tiny, even 'crowding out' public schooling. Evidence together: moderate effects in small samples (d≈0.42–0.67), but a 2024 meta-analysis of 265 RCTs finds at policy-relevant scale the effect is only 1/3–1/2 of small-study estimates; NBER ~0.37 SD. UNESCO now calls private tutoring a global 'shadow education.'

◆ 杠杆排序 · 三份研究一字不差地收敛于此（结论卡）The leverage order · all three studies converge here

1亲子关系与语言环境＞ ~~课程数量~~Parent-child bond & language environment ＞ ~~number of courses~~

2选赛道＞ ~~赛道内加码~~Choosing the track ＞ ~~piling on within the track~~

3家长自身状态＞ ~~对孩子的干预强度~~The parent's own state ＞ ~~intensity of intervention on the child~~

理由：亲子 / 语言环境是少数 AI 无法替代且回报持久的投入；选轨的最大变量常是制度 / 政策风险而非机构水平；家长焦虑会直接传导并损害孩子的内生动机。Why: the bond / language environment is one of the few investments AI can't replace and whose returns last; the biggest variable in choosing a track is usually institutional / policy risk, not vendor quality; parental anxiety transmits directly and damages the child's intrinsic motivation.

从这张图看到的五条规律Five patterns this map makes visible